Type Inference

Boo is a statically typed language.

Static typing is about the ability to type check a program for type correctness.

Static typing is about being able to deliver better runtime performance.

Static typing is not about putting the burden of declaring types on the programmer as most mainstream languages do.

The mechanism that frees the programmer from having to babysit the compiler is called type inference.

Type inference means you don't have to worry about declaring types everywhere just to make the compiler happy. Type inference means you can be productive without giving up the safety net of the type system nor sacrificing performance.

Boo's type inference kicks in for newly declared variables and fields, properties, arrays, for statement variables, overriden methods, method return types and generators.

Variables

Assignments can be used to introduce new variables in the current scope. The type for the new variable will be inferred from the expression on the right.

s1 = "foo" # declare new variable s1
s2 = s1.Replace("f", "b") # s1 is a string so Replace is cool

Only the first assignment to a variable is taken into account by the type inference mechanism.
The following program is illegal:

s = "I'm a string" # s is bound with type string
s = 42 # and although 42 is a really cool number s can only hold strings

Fields

class Customer:

    _name = ""

Declare the new field _name and initialize it with an empty string. The type of the field will be string.

Properties

When a property does not declare a type it is inferred from its getter.

class Customer:

    _name = ""

    Name:
        get:
            return _name

In this case the type of the Name property will be inferred as string.

Arrays

The type of an array is inferred as the [least generic type] that could safely hold all of its enclosing elements.

a = (1, 2) # a is of type (int)

b = (1L, 2) # b is of type (long)

c = ("foo", 2) # c is of type (object)

For statement variables

names = (" John ",
         " Eric",
         "  Graham",
         "TerryG  ",
         " TerryJ",
         " Michael")

for name in names:
    # name is typed string since we are iterating a string array
    print(name.Trim()) # Trim is cool, name is a string

This works even when with unpacking:

a = ( (1, 2), (3, 4) )

for i, j in a:
     print(i+j) # + is cool since i and j are typed int

Overriden methods

When overriding a method is not necessary to declare its return type since it can be safely inferred from its super method.

class Customer:

    override def ToString():
         pass

Method return types

The return type of a method will the [most generic type] among the types of the expressions used in return statements.

def spam():
    return "spam!"

print(spam()*3)
# multiply operator is cool since spam() is inferred to return a string
# and strings can be multiplied by integer values

def ltuae(returnString as bool):
    return "42" if returnString
    return 42

# ltuae is inferred to return object
print(ltuae(false)*3) # ERROR! don't know the meaning of the * operator

When a method does not declare a return type and includes no return statements it will be typed System.Void.

Generators

g = i*2 for i in range(3)

# g is inferred to generate ints

for i in g:
    print(i*2) # * operator is cool since i is inferred to be int

# works with arrays too
a = array(g) # a is inferred to be (int) since g delivers ints

print(a[0] + a[-1]) # int sum

A Word of Caution About Interfaces

When implementing interfaces it's important to explicitliy declare the signature of a method, property or event. The compiler will look only for exact matches.

In the example below the class will be considered abstract since it does not provide an implementation with the correct signature:

namespace AllThroughTheDay

interface IMeMineIMeMineIMeMine:

	def AllThroughTheNight(iMeMine, iMeMine, iMeMine as int)

class EvenThoseTears(IMeMineIMeMineIMeMine):

	def AllThroughTheNight(iMeMine, iMeMine, iMeMine):
		pass

e = EvenThoseTears()

Ok. So where I do have to declare types then?

Let's say when.

• when the compiler as it exists today can't do it for you. Ex: parameter types, recursive and mutually recursive method/property/field definitions, return for abstract and interface methods, for in untyped containers, properties with a only set defined

  abstract def Method(param /* as object */, i as int) as string:
      pass
  
  def fat([required(value >= 0)] value as int) as int:
      return 1 if value < 2
      return value*fat(value-1)
  
  for i as int in [1, 2, 3]: # list is not typed
      print(i*2)

• when you don't want to express what the compiler thinks you do:

  def foo() as object:  # I want the return type to be object not string
      # a common scenario is interface implementation
      return "a string"
  
  if bar:
      a = 3 # a will be typed string
  else:
      a = "42" # uh, oh

• when you want to access a member not exposed by the type assigned to an expression:

  f as IDisposable = foo()
  f.Dispose()

• when you want to use Duck Typing:

  ie as duck = CreateInstance("InternetExplorer.Application")
  ie.Visible = true
  ie.Navigate("http://boo.codehaus.org/Type+Inference")